We present SHIFT3D, a differentiable pipeline for generating 3D shapes that are structurally plausible yet challenging to 3D object detectors. In safety-critical applications like autonomous driving, discovering such novel challenging objects can offer insight into unknown vulnerabilities of 3D detectors. By representing objects with a signed distanced function (SDF), we show that gradient error signals allow us to smoothly deform the shape or pose of a 3D object in order to confuse a downstream 3D detector. Importantly, the objects generated by SHIFT3D physically differ from the baseline object yet retain a semantically recognizable shape. Our approach provides interpretable failure modes for modern 3D object detectors, and can aid in preemptive discovery of potential safety risks within 3D perception systems before these risks become critical failures.

With recent advances in speech synthesis, synthetic data is becoming a viable alternative to real data for training speech recognition models. However, machine learning with synthetic data is not trivial due to the gap between the synthetic and the real data distributions. Synthetic datasets may contain artifacts that do not exist in real data such as structured noise, content errors, or unrealistic speaking styles. Moreover, the synthesis process may introduce a bias due to uneven sampling of the data manifold. We propose two novel techniques during training to mitigate the problems due to the distribution gap: (i) a rejection sampling algorithm and (ii) using separate batch normalization statistics for the real and the synthetic samples. We show that these methods significantly improve the training of speech recognition models using synthetic data. We evaluate the proposed approach on keyword detection and Automatic Speech Recognition (ASR) tasks, and observe up to 18% and 13% relative error reduction, respectively, compared to naively using the synthetic data.

Controllable generative sequence models with the capability to extract and replicate the style of specific examples enable many applications, including narrating audiobooks in different voices, auto-completing and auto-correcting written handwriting, and generating missing training samples for downstream recognition tasks. However, under an unsupervised-style setting, typical training algorithms for controllable sequence generative models suffer from the training-inference mismatch, where the same sample is used as content and style input during training but unpaired samples are given during inference. In this paper, we tackle the training-inference mismatch encountered during unsupervised learning of controllable generative sequence models. The proposed method is simple yet effective, where we use a style transformation module to transfer target style information into an unrelated style input. This method enables training using unpaired content and style samples and thereby mitigate the training-inference mismatch. We apply style equalization to text-to-speech and text-to-handwriting synthesis on three datasets. We conduct thorough evaluation, including both quantitative and qualitative user studies. Our results show that by mitigating the training-inference mismatch with the proposed style equalization, we achieve style replication scores comparable to real data in our user studies.

Data augmentation methods usually apply the same augmentation (or a mix of them) to all the training samples. For example, to perturb data with noise, the noise is sampled from a Normal distribution with a fixed standard deviation, for all samples. We hypothesize that a hard sample with high training loss already provides strong training signal to update the model parameters and should be perturbed with mild or no augmentation. Perturbing a hard sample with a strong augmentation may also make it too hard to learn from. Furthermore, a sample with low training loss should be perturbed by a stronger augmentation to provide more robustness to a variety of conditions. To formalize these intuitions, we propose a novel method to learn a Sample-Adaptive Policy for Augmentation – SapAugment. Our policy adapts the augmentation parameters based on the training loss of the data samples. In the example of Gaussian noise, a hard sample will be perturbed with a low variance noise and an easy sample with a high variance noise. Furthermore, the proposed method combines multiple augmentation methods into a methodical policy learning framework and obviates hand-crafting augmentation parameters by trial-and-error. We apply our method on an automatic speech recognition (ASR) task, and combine existing and novel augmentations using the proposed framework. We show substantial improvement, up to 21% relative reduction in word error rate on LibriSpeech dataset, over the state-of-the-art speech augmentation method.

Deep Neural Network–Hidden Markov Model (DNN-HMM) based methods have been successfully used for many always-on keyword spotting algorithms that detect a wake word to trigger a device. The DNN predicts the state probabilities of a given speech frame, while HMM decoder combines the DNN predictions of multiple speech frames to compute the keyword detection score. The DNN, in prior methods, is trained independent of the HMM parameters to minimize the cross-entropy loss between the predicted and the ground-truth state probabilities. The mismatch between the DNN training loss (cross-entropy) and the end metric (detection score) is the main source of sub-optimal performance for the keyword spotting task. We address this loss-metric mismatch with a novel end-to-end training strategy that learns the DNN parameters by optimizing for the detection score. To this end, we make the HMM decoder (dynamic programming) differentiable and back-propagate through it to maximize the score for the keyword and minimize the scores for non-keyword speech segments. Our method does not require any change in the model architecture or the inference framework; there-fore, there is no overhead in run-time memory or compute requirements. Moreover, we show significant reduction in false rejection rate (FRR) at the same false trigger experience (> 70% over independent DNN training).

We present a method to generate speech from input text and a style vector that is extracted from a reference speech signal in an unsupervised manner, i.e., no style annotation, such as speaker information, is required. Existing unsupervised methods, during training, generate speech by computing style from the corresponding ground truth sample and use a decoder to combine the style vector with the input text. Training the model in such a way leaks content information into the style vector. The decoder can use the leaked content and ignore some of the input text to minimize the reconstruction loss. At inference time, when the reference speech does not match the content input, the output may not contain all of the content of the input text. We refer to this problem as "content leakage", which we address by explicitly estimating and minimizing the mutual information between the style and the content through an adversarial training formulation. We call our method MIST - Mutual Information based Style Content Separation. The main goal of the method is to preserve the input content in the synthesized speech signal, which we measure by the word error rate (WER) and show substantial improvements over state-of-the-art unsupervised speech synthesis methods.

With recent progress in graphics, it has become more tractable to train models on synthetic images, potentially avoiding the need for expensive annotations. However, learning from synthetic images may not achieve the desired performance due to a gap between synthetic and real image distributions. To reduce this gap, we propose Simulated+Unsupervised (S+U) learning, where the task is to learn a model to improve the realism of a simulators output using unlabeled real data, while preserving the annotation information from the simulator. We develop a method for S+U learning that uses an adversarial network similar to Generative Adversarial Networks (GANs), but with synthetic images as inputs instead of random vectors. We make several key modifications to the standard GAN algorithm to preserve annotations, avoid artifacts, and stabilize training: (i) a self-regularization term, (ii) a local adversarial loss, and (iii) updating the discriminator using a history of refined images. We show that this enables generation of highly realistic images, which we demonstrate both qualitatively and with a user study. We quantitatively evaluate the generated images by training models for gaze estimation and hand pose estimation. We show a significant improvement over using synthetic images, and achieve state-of-the-art results on the MPIIGaze dataset without any labeled real data.

Many existing recognition algorithms combine different modalities based on training accuracy but do not consider the possibility of noise at test time. We describe an algorithm that perturbs test features so that all modalities predict the same class. We enforce this perturbation to be as small as possible via a quadratic program (QP) for continuous features, and a mixed integer program (MIP) for binary features. To efficiently solve the MIP, we provide a greedy algorithm and empirically show that its solution is very close to that of a state-of-the-art MIP solver. We evaluate our algorithm on several datasets and show that the method outperforms existing approaches.

We present a multi-class, multiple instance learning (MIL) algorithm using the dictionary learning framework where the data is given in the form of bags. Each bag contains multiple samples, called instances, out of which at least one belongs to the class of the bag. We propose a noisy-OR model-based optimization framework for learning the dictionaries. Our method can be viewed as a generalized dictionary learning algorithm since it reduces to a novel discriminative dictionary learning framework when there is only one instance in each bag. Various experiments using the popular MIL datasets show that the proposed method performs better than existing methods.

When designing classifiers for classification tasks, one is often confronted with situations where data distributions in the source domain are different from those present in the target domain. This problem of domain adaptation is an important problem that has received a lot of attention in recent years. In this paper, we study the challenging problem of unsupervised domain adaptation, where no labels are available in the target domain. In contrast to earlier works, which assume a single domain shift between the source and target domains, we allow for multiple domain shifts. Towards this, we develop a novel framework based on the parallel transport of union of the source subspaces on the Grassmann manifold. Various recognition experiments show that this way of modeling data with union of subspaces instead of a single subspace improves the recognition performance.

In this paper, we propose a multiple kernel learning (MKL) algorithm that is based on the sparse representation-based classification (SRC) method. Taking advantage of the nonlinear kernel SRC in efficiently representing the nonlinearities in the high-dimensional feature space, we propose an MKL method based on the kernel alignment criteria. Our method uses a two step training method to learn the kernel weights and sparse codes. At each iteration, the sparse codes are updated first while fixing the kernel mixing coefficients, and then the kernel mixing coefficients are updated while fixing the sparse codes. These two steps are repeated until a stopping criteria is met. The effectiveness of the proposed method is demonstrated using several publicly available image classification databases and it is shown that this method can perform significantly better than many competitive image classification algorithms.

While recent techniques for discriminative dictionary learning have demonstrated tremendous success in image analysis applications, their performance is often limited by the amount of labeled data available for training. Even though labeling images is difficult, it is relatively easy to collect unlabeled images either by querying the web or from public datasets. In this paper, we propose a discriminative dictionary learning technique which utilizes both labeled and unlabeled data for learning dictionaries. Extensive evaluation on existing datasets demonstrate that the proposed method performs significantly better than state of the art dictionary learning approaches when unlabeled images are available for training.

Segmenting “simple” objects using low-level visual cues is an important capability for a vision system to learn in an unsupervised manner. We define a “simple” object as a compact region enclosed by depth and/or contact boundary in the scene. We propose a segmentation process to extract all the “simple” objects that builds on the fixation-based segmentation framework [1] that segments a region given a point anywhere inside it. In this work, we augment that framework with a fixation strategy to automatically select points inside the “simple” objects and a post-segmentation process to select only the regions corresponding to the “simple” objects in the scene. A novel characteristic of our approach is the incorporation of border ownership, the knowledge about the object side of a boundary pixel. We evaluate the process on a publicly available RGB-D dataset [2] and find that the proposed method successfully extracts 91.4% of all objects in the dataset.